Estimation of Various Human Emotions Using Lightweight Fnirs Device

Abstract

We previously proposed a method for estimating pleasant and unpleasant emotions with high accuracy using only total hemoglobin data measured with a lightweight functional near infrared spectroscopy device. In this study, we used the method to evaluate the accuracy of estimating 20 types of emotions selected as uniformly distributed emotions in Russell's circumplex model. We first divided the 20 types of emotions into four groups, corresponding to the four quadrants of Russell's circumplex model and evaluated the estimation accuracy of each quadrant. The results indicate that the activation quadrant was estimated with high accuracy when the emotion was strongly aroused, with 76.7% recall for the pleasant-activation quadrant and 72.2% recall for the unpleasant-activation quadrant. We then evaluated the estimation accuracy of the 20 emotions individually. The results indicate that "excited" and "lethargic" were estimated with high accuracy, with 73.3% recall for "excited" and 61.5% recall for "lethargic," and recall of "excited" improved to 80% when the emotion was strongly aroused. The results of this study indicate that the more strongly emotions included in activation quadrant in Russell's circumplex model are aroused, the more accurately they can be classified. "Excited" and "lethargic" could be estimated with high accuracy regardless of the degree of emotional arousal.